Video motion magnification (VMM) allows to amplify hardly visible changes in the video input sequence. It is possible to amplify breathing motion of hospital patients or oscillations of some mechanical element without need to connect measuring devices. On the other hand modified VMM algorithm can be used to extinguish motion before processing with another method. Unfortunately, algorithm in its full, but slowest version uses new pyramids instead laplacian ones and requires remarkable computational power in each stage of processing and that problem is solved by GPU-based implementation of basic operations using CUDA technology and scheduling them by CPU. That approach lets to run all computations on GPU. Additionally in comparison with previous studies temporal filter was changed to butterworth one. Although testing hardware was only able to run low resolution in real time, there are no doubts that better GPU would be able to run 640 × 480 resolution in real time.
展开▼